OpenInference Instrumentation
Utility functions for OpenInference instrumentation.
Installation
pip install openinference-instrumentation
Annotation and Evaluation Attributes
Use get_annotation_attributes and get_evaluation_attributes to turn typed
Annotation objects into flattened OpenInference span attributes. Both helpers
support "span" (the default), "trace", and "session" scopes and assign
contiguous collection indices in input order.
from openinference.instrumentation import (
Annotation,
get_annotation_attributes,
get_evaluation_attributes,
)
span_annotations = get_annotation_attributes(
annotations=[
Annotation(
name="hallucination",
label="factual",
explanation="Every claim is supported by the retrieved documents.",
annotator_kind="LLM",
identifier="judge-v2",
metadata={"rubric_version": 2},
)
]
)
trace_evaluations = get_evaluation_attributes(
evaluations=[Annotation(name="correctness", score=0.95)],
scope="trace",
)
span.set_attributes({**span_annotations, **trace_evaluations})
Annotation has these fields:
name(required): criterion or metric name.- At least one of
score,label, orexplanation(required by the helpers). annotator_kind: conventionally"HUMAN","LLM", or"CODE"; custom values are allowed.identifier: stable producer-assigned result identifier.metadata: a dictionary to JSON-serialize, or an already serialized JSON object string.
The evaluation helper uses the same Annotation model because evaluation is an
alternative attribute terminology for annotations. It emits evaluations.*
instead of annotations.*. Trace and session scopes add the corresponding
trace. or session. prefix. Session-scoped annotations also require the
carrying span to have session.id; post-hoc span and trace annotations require
the target Span Link described in the
annotation specification.
Customizing Spans
The openinference-instrumentation package offers utilities to track important application metadata such as sessions and metadata using Python context managers:
using_session: to specify a session ID to track and group a multi-turn conversation with a userusing_user: to specify a user ID to track different conversations with a given userusing_metadata: to add custom metadata, that can provide extra information that supports a wide range of operational needsusing_tag: to add tags, to help filter on specific keywordsusing_prompt_template: to reflect the prompt template used, with its version and variables. This is useful for prompt template managementusing_attributes: it helps handling multiple of the previous options at once in a concise manner
For example:
from openinference.instrumentation import using_attributes
tags = ["business_critical", "simple", ...]
metadata = {
"country": "United States",
"topic":"weather",
...
}
prompt_template = "Please describe the weather forecast for {city} on {date}"
prompt_template_variables = {"city": "Johannesburg", "date":"July 11"}
prompt_template_version = "v1.0"
with using_attributes(
session_id="my-session-id",
user_id="my-user-id",
metadata=metadata,
tags=tags,
prompt_template=prompt_template,
prompt_template_version=prompt_template_version,
prompt_template_variables=prompt_template_variables,
):
# Calls within this block will generate spans with the attributes:
# "session.id" = "my-session-id"
# "user.id" = "my-user-id"
# "metadata" = "{\"key-1\": value_1, \"key-2\": value_2, ... }" # JSON serialized
# "tag.tags" = "["tag_1","tag_2",...]"
# "llm.prompt_template.template" = "Please describe the weather forecast for {city} on {date}"
# "llm.prompt_template.variables" = "{\"city\": \"Johannesburg\", \"date\": \"July 11\"}" # JSON serialized
# "llm.prompt_template.version " = "v1.0"
...
Each helper also works as a decorator. The attributes stay attached for the whole call,
including across await suspension points of async def functions and while the body of a
generator or async def generator runs (without leaking into the code that consumes it):
from openinference.instrumentation import using_session, using_user
@using_session("my-session-id")
@using_user("my-user-id")
async def answer(question: str) -> str:
# Spans created here, and by any awaited instrumented call, carry
# "session.id" = "my-session-id" and "user.id" = "my-user-id"
...
When combining them with span-creating decorators such as tracer.agent or tracer.tool, put
the using_* decorators on top: the attributes are copied onto a span when it starts, so they
have to be attached before the span-creating decorator runs.
See examples/async_context_attribute_decorators.py for a runnable example that exports to a local Phoenix server.
You can read more about this in our docs.
Suppressing Tracing
suppress_tracing turns off span creation for everything inside its block. It works as a
regular context manager and as an async one:
from openinference.instrumentation import suppress_tracing
with suppress_tracing():
... # no spans are created here
async with suppress_tracing():
... # nor here, including across awaits
See examples/suppress_tracing_sync_async.py for a runnable example that exports to a local Phoenix server and shows both forms.
Tracing Configuration
This package contains the central TraceConfig class, which lets you specify a tracing configuration that lets you control settings like data privacy and payload sizes. For instance, you may want to keep sensitive information from being logged for security reasons, or you may want to limit the size of the base64 encoded images logged to reduced payload size.
In addition, you an also use environment variables, read more here. The following is an example of using the TraceConfig object:
from openinference.instrumentation import TraceConfig
config = TraceConfig(
hide_inputs=hide_inputs,
hide_outputs=hide_outputs,
hide_input_messages=hide_input_messages,
hide_output_messages=hide_output_messages,
hide_input_images=hide_input_images,
hide_input_text=hide_input_text,
hide_output_text=hide_output_text,
base64_image_max_length=base64_image_max_length,
)
tracer_provider = ...
# This example uses the OpenAIInstrumentor, but it works with any of our auto instrumentors
OpenAIInstrumentor().instrument(tracer_provider=tracer_provider, config=config)
1 maintainer
Metadata
1 maintainer
Release files for openinference-instrumentation 0.1.72
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| openinference_instrumentation-0.1.72.tar.gz | 46.5 kB | With provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| openinference_instrumentation-0.1.72-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 101.0 kB
Release files / openinference_instrumentation-0.1.72.tar.gz
| Download URL | openinference_instrumentation-0.1.72.tar.gz |
|---|---|
| Size | 46.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
4470a316dcb279aff45c2368e16d76ec43b2f46985ee2a4dab183b448c1bc18b
|
|
BLAKE2b-256 checksum How to use checksums |
555d281d31ef653257875209cdaef355157bf4ca5e99599df88c3e02ccdbd4a7
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 23, 2026.
Transparency logRelease files / openinference_instrumentation-0.1.72-py3-none-any.whl
| Download URL | openinference_instrumentation-0.1.72-py3-none-any.whl |
|---|---|
| Size | 54.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
1eaba9604e6a0de4829094f2b5f6796a3962c36c72752b1ac7c01fca27540a0f
|
|
BLAKE2b-256 checksum How to use checksums |
d2bed63a820efd759f04d6463380807e1dd04ce898d54cddec23e39ef5abd216
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 23, 2026.
Transparency logSecurity for openinference-instrumentation 0.1.72
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
All 2 files in this release have been uploaded with provenance metadata.
Attestation for openinference_instrumentation-0.1.72.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 23, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
